Transformer Fault Diagnosis Model Based on Iterative Nearest Neighbor Interpolation and Ensemble Learning

Abstract

In order to improve the accuracy of fault diagnosis of power transformer in the presence of missing values, we propose a missing value imputation method by iterative k nearest neighbor (KNN), and ensemble learning is used to diagnose the power transformer. First, according to importance of different attributes of dissolved gas data in the ensemble classification method, interpolation order of missing data is determined. Then, in order to take advantage of valid information from data with missing values, the missing values are interpolated in an iterative manner by KNN. Finally, the ensemble learning model is used for fault diagnosis of transformers. The experimental results show that the average accuracy of transformer fault diagnosis after using this method to interpolate DGA data sets is increased by 15.4%, and the average accuracy can reach 82%.